KR101771855B1 - Method for calculating relative trust information of recommendation system and method for recommending item based on trust - Google Patents

Abstract

Disclosed is a method for calculating relative reliability, which is based on a practical reliability relation and improves a tendency problem on a grade, and a method for recommending an item based on reliability. The method for calculating the relative reliability includes: a step of calculating the relative reliability between two users by using an average grade on an item and the grade of the two users on the at least an item; a step of generating a reliability network having the multiple users including the two users as a node by connecting edges between the nodes in which the relative reliability exists; and a step of estimating the relative reliability between the nodes in which the relative reliability does not exist, by using the calculated relative reliability.

Description

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a method for calculating relative reliability for a recommendation system and a method for recommending an item based on trust,

The present invention relates to a relative reliability calculation method and a trust based item recommendation method for a recommendation system.

The most prominent method of recommending items such as contents and products is a method using collaborative filtering developed at the University of Minnesota since the 1990s.

A collaborative filtering technique is a method of predicting a user's interest based on preference information obtained from a user. More specifically, the user's preference is predicted based on the preference of the similar user group, provided that the user having a similar tendency prefers a similar item.

However, there is a problem of information sparse occurring when there is insufficient data, a coldstart problem that occurs when there is no information about a user or a product, Opinion spam deliberately damaging or promoting the reputation of a specific product, , And the disadvantages such as the tendency of the rating tendency were raised.

To improve this, a trust - based recommendation method has been proposed in which the concept of social network service is applied to item recommendation method, and users can directly select users they trust. This trust-based recommendation method is based on the premise that users tend to prefer products recommended by friends who are more trustworthy than strangers.

The reliability of the trust - based recommendation method is divided into two types of explicit reliability and implicit reliability. Explicit reliability is the reliability information that the user registers directly with the system, and implicit reliability is the reliability information deduced from the rating data.

In the case of item recommendation method based on explicit reliability, it is more likely that users will not disclose the reliability of each other. Therefore, it is more likely that information sparseness problem occurs than rating data. A tendency problem may occur. Recently, research on recommending items based on tacit reliability has been actively carried out.

However, since the conventional method based on the implicit reliability considers the ratings of the users having different rating criteria, the tendency problem still exists, and the relative reliability of the user is symmetrically expressed to reflect the asymmetric trust relationship of the reality There is a problem that can not be done.

A related prior art document is Korean Patent No. 10-1572719.

The present invention is to provide a relative reliability calculation method and a trust-based item recommendation method for a recommendation system.

In particular, the present invention is to provide an item recommendation method for improving a tendency problem on a rating and reflecting a more realistic trust relationship, and a relative reliability calculation method for item recommendation.

According to an embodiment of the present invention, there is provided a method of calculating relative reliability between two users using two users' ratings of at least one item and an average rating of the items. Connecting an edge between the nodes where the relative reliability exists and creating a trust network having a plurality of users including the two users as the nodes; And estimating a relative reliability between the nodes for which the relative reliability does not exist, using the calculated relative reliability.

According to another embodiment of the present invention, there is provided a method of calculating relative reliability between two users using two users' ratings of at least one item and an average rating of the items, Estimating a relative reliability among users whose relative reliability does not exist using the calculated relative reliability; And recommending an item for a target user using the relative reliability.

According to another aspect of the present invention, there is provided a method of providing a user interface, the method comprising: receiving ratings of users for at least one item; Calculating relative reliability between users who have given a common rating to a particular item; And recommending to the target user a target user and an item preferred by the user with a higher relative reliability, wherein the step of calculating the relative reliability comprises calculating a user's rating for the particular item and an average rating for the item Based item recommendation method for calculating the relative reliability between two users.

According to the present invention, it is possible to recommend items satisfactory to a specific user by recommending items preferred by target users to a specific user, using item preferences of target users with high relative reliability to a specific user.

Further, according to the present invention, the tendency problem can be improved by reflecting the average rating of the item on the relative preference.

1 is a view for explaining a trust-based item recommendation apparatus according to an embodiment of the present invention.
2 is a flowchart illustrating a method of calculating a relative reliability for a recommendation system according to an embodiment of the present invention.
3 is a diagram for explaining a relative reliability and trust network according to the present invention.
4 is a diagram for explaining a trust-based item recommendation method according to an embodiment of the present invention.
5 is a view for explaining a trust-based item recommendation method according to another embodiment of the present invention.

While the invention is susceptible to various modifications and alternative forms, specific embodiments thereof are shown by way of example in the drawings and will herein be described in detail. It should be understood, however, that the invention is not intended to be limited to the particular embodiments, but includes all modifications, equivalents, and alternatives falling within the spirit and scope of the invention. Like reference numerals are used for like elements in describing each drawing.

The present invention relates to an item recommendation method based on implicit reliability, which calculates relative reliability between users and recommends an item using the relative reliability. Here, the item may include items such as a movie or a game in which a rating is left, an internet service that shares a picture or the like, an investment object, and the like, as well as items that can be purchased.

Relative reliability is calculated among users who have been rated for the same item. Relative reliability is defined between two users, and two relative reliability is calculated between two users because one side trusts the other. For example, the relative reliability can be divided into a first relative reliability of the first user to the second user and a second relative reliability of the two users to the first user.

The relative reliability according to the present invention is determined according to the rating difference between the two users and is inversely proportional to the user's rating difference. That is, if the difference between the two users' ratings is large, the relative reliability is low. If the difference between the two users is small, the relative reliability is high. Accordingly, when the first relative reliability is high, the influence of the second user on the first user is high when the first user purchases the article, and when the second relative reliability is high, the influence of the first user on the second user Is high.

As a result, the present invention can recommend an item satisfactory to a specific user by recommending a favorite item of a target user to a specific user, using the item preference of a target user with a relatively high reliability.

Meanwhile, the present invention calculates a relative reliability by using an average score of an item to improve a tendency problem on a rating.

For example, when the user rating of a tendency to leave behind a good rating regardless of the actual quality of the item or the satisfaction of the item is used as it is, there is a problem that the relative reliability is not accurate. The tendency problem can be improved by reflecting the average rating to the relative reliability and reducing the influence on the user's rating.

Hereinafter, embodiments according to the present invention will be described in detail with reference to the accompanying drawings.

1 is a view for explaining a trust-based item recommendation apparatus according to an embodiment of the present invention.

Referring to FIG. 1, the trust-based item recommendation apparatus according to the present invention includes a relative reliability calculation unit 110 and an item recommendation unit 120. The trust-based item recommendation apparatus according to the present invention may be employed in a server of an Internet shopping mall or a platform company that sells and recommends items as an embodiment.

The relative reliability calculation unit 110 calculates a relative reliability between users who have given a common rating to a specific item. The relative reliability calculation unit 110 may calculate the relative reliability between the users by receiving ratings of the users after purchasing or using the items. If there are more than two users who purchase a particular item, the relative reliability between two different users can be calculated.

At this time, the relative reliability calculation unit 110 calculates the relative reliability between the two users, using the user's rating for the specific item and the average rating for the item. Relative reliability can be increased as the two users' ratings are similar, and the relative reliability can be reduced as the difference between the user's rating and the average rating is greater.

The item recommendation unit 120 recommends a target user and an item preferred by the user having a relatively high reliability to the target user, using the calculated relative reliability.

On the other hand, according to the embodiment, the relative reliability calculation unit 110 can estimate the relative reliability between users having no relative reliability and can use a trust network generated according to the relative reliability calculated for the relative reliability prediction.

In addition, the item recommending unit 120 may recommend a target user's favorite item to the target user in consideration of the calculated relative reliability as well as the predicted relative reliability.

Hereinafter, the relative reliability calculation method for the recommendation system will be described first, and the trust based item recommendation method using the calculated relative reliability will be described.

FIG. 2 is a flowchart illustrating a relative reliability calculation method for a recommendation system according to an embodiment of the present invention. FIG. 3 is a diagram for explaining a relative reliability and trust network according to the present invention.

The relative reliability calculating method according to the present invention can be performed in a computing device including the above-described relative reliability calculating unit or processor.

The relative reliability extraction apparatus according to the present invention calculates the relative reliability between two users (S210) using the ratings of two users for at least one item and the average rating for the items. More specifically, the relative reliability extracting apparatus can calculate the relative reliability by using the difference between the ratings of the two users and the difference between the user's rating and the average rating.

As described above, two relative reliability is calculated between two users, and the relative reliability extracting device calculates a first relative reliability for the second user of the first user, a second relative reliability for the first user of the second user Can be calculated.

As an example, the relative reliability extracting apparatus calculates a first relative reliability for a first user (user u) of a second user (user v) among the relative reliability of two users for an individual item using Equation (1) If two items are purchased in common by two users, the calculation result may be averaged over all the items having the ratings of the two users as shown in Equation (2), and finally the first relative reliability may be calculated.

는 r_u,i 및 r_v,i를 제외한 아이템 i에 대한 평점의 개수,

는 는 r_u,i 및 r_v,i를 제외한 아이템 i에 대한 평균 평점,

는 사용자 u와 v가 공통으로 평점을 남긴 아이템 개수를 나타낸다.Here, r _max is the maximum rating that the user can give, r _{u, i} is the rating given by user u to item i, r _{v, i} is the rating given to item i by user v,

Is the number of ratings for item i, excluding r _{u, i} and r _{v, i} ,

Is the average rating for item i excluding r _{u, i} and r _{v, i} ,

Represents the number of items in which users u and v have a common rating.

According to Equation (1), the relative reliability increases as the two users' ratings are similar, and the relative reliability decreases as the difference between the user and the average rating increases. The average rating used in Equation (1) may be an average rating for users other than the two users (u, v), or an average rating including two users (u, v) depending on the embodiment.

(1) is a mathematical expression for the first relative reliability in which the user u trusts the user v, and when the subscripts u and v are mutually exchanged in Equation 1, the user v (the second user) A second relative reliability can be calculated.

The relative reliability extracting apparatus according to the present invention generates a trust network using relative reliability (S210), connects the edges between nodes having relative reliability, and transmits a trust network having a plurality of users including the two users as nodes . 3 (a), the trust network as shown in FIG. 3 (b) can be created.

In FIG. 3A, a line connecting the users 310 to 350 with the items A, B, C, and D corresponds to a rating left after purchasing the item. In the trust network of FIG. 3 (b), each node corresponds to a user, and the edge connecting the nodes indicates that there is a relative reliability. The direction of the edge indicates the directionality according to the first relative reliability and the second relative reliability.

Since the first user 310, the second user 320 and the fifth user 350 have the items B and C purchased in common, the relative reliability can be calculated. Since the third to fifth users 330, 340, and 350 also have the items A purchased in common, the relative reliability can be calculated.

On the other hand, since the first user 310 does not have an item purchased in common with the third user 330 and there is no item purchased in common with the fourth user 340, the first and third users 310 and 330 The relative reliability between the first and fourth users 310 and 340 can not be calculated. Similarly, since the second user 320 also has no items purchased in common with the third and fourth users 330 and 340, the relative reliability between the second and third users 320 and 330, 0.0 > 320, < / RTI > 340 can not be calculated.

As a result, if users buy an item and leave a rating as shown in FIG. 3 (a), a trust network as shown in FIG. 3 (b) can be finally created and a relative reliability can be assigned to an edge of the trust network.

Referring back to FIG. 2, the relative reliability extracting apparatus estimates the relative reliability between nodes having no relative reliability using the calculated relative reliability (S230). That is, the relative reliability extracting apparatus predicts the relative reliability among the users having no relative reliability.

The relative reliability extraction apparatus determines the shortest path between the first node and the second node in which the relative reliability does not exist, and weights the relative reliability of the edge included in the shortest path to predict the relative reliability.

Referring to FIG. 3, for example, there is no relative reliability between the first user node 310 and the third user node 330 in FIG. 3 (b). The shortest path among the various paths from the first user node 310 to the third user node 330 is a path via only the fifth user node 350 without passing through another node. The shortest path includes two edges from the first user node 310 toward the third user node 330. [ Accordingly, the relative reliability extraction apparatus applies a weight to the relative reliability between the first user node 310 and the fifth user node 350 and the relative reliability between the fifth user node 350 and the third user node 330, respectively Thereby estimating the relative reliability between the first user node 310 and the third user node 330. [

At this time, the weights can be adjusted according to the distance between the first node and the edge as the reference, and larger weights can be set at the edge near the first node. That is, the weight of the edge far from the reference node becomes smaller. This can be expressed by the following equation (3).

The relative reliability predicted according to Equation (3) can be normalized to the median of all weights applied to the prediction.

는 가중치이며,

는 최단 경로에 포함된 에지에 할당된 상대적 신뢰도이다. 그리고 k는 최단 경로에 포함되는 에지의 개수, i는 에지의 인덱스로서 기준노드에서 멀어지는 방향으로 1씩 증가한다.here,

Is a weight,

Is the relative reliability assigned to the edges included in the shortest path. K is the number of edges included in the shortest path, and i is an index of the edge, which increases by 1 in the direction away from the reference node.

When the relative reliability that the first user 310 trusts the third user 330 is predicted, the reference node becomes the first user node 310 and the third user 330 becomes the first user 310 The reference node may be the third user node 330 when a reliable relative reliability is predicted.

On the other hand, when the trust network is complicated and the shortest path is long, the complexity of the calculation may increase. Therefore, the relative reliability extracting apparatus does not predict relative relative roadway using all the relative reliability included in the shortest path, The relative reliability can be predicted by applying a weight to the relative reliability of edges existing below the critical distance.

As a result, according to the present invention, the tendency problem can be improved by reflecting the average rating for the item in the relative reliability, and by predicting the relative reliability between users having no relative reliability, can do.

Also, according to the present invention, the relative reliability between two users represents an asymmetric structure characteristic that varies according to a rating given to each user, and the relative reliability can be expressed more realistically.

4 is a diagram for explaining a trust-based item recommendation method according to an embodiment of the present invention.

The trust-based item recommendation method according to the present invention may be performed in a computing device including the above-described trust-based item recommendation device or processor.

The trust-based item recommendation apparatus according to the present invention calculates the relative reliability between the two users using the ratings of the two users for at least one item and the average rating for the items (S410). Then, using the calculated relative reliability, the relative reliability between users having no relative reliability is estimated (S420).

That is, the trust-based item recommendation apparatus according to the present invention can calculate the relative reliability as described with reference to FIG. Then, as described in FIG. 2, a trust network can be created, and the relative reliability between nodes without relative reliability can be predicted.

Based on the computed and predicted relative reliability, the trust-based item recommendation apparatus according to the present invention recommends an item for the target user (S430).

For example, the trust-based item recommending apparatus may select a predetermined number of users in order of higher relative reliability with respect to the target user, and recommend an item that the selected users prefer, for example, a user with a high rating, to the target user.

5 is a view for explaining a trust-based item recommendation method according to another embodiment of the present invention.

The trust-based item recommendation apparatus according to the present invention receives user ratings of at least one item created by users and calculates a relative reliability between users who have given a common rating to a specific item (S510). Then, an item preferred by the user having a high relative reliability with the target user is recommended to the target user (S520).

Compared to FIG. 4, the item recommendation method of FIG. 5 recommends an item to a target user based on relative reliability calculated without relative reliability prediction.

The above-described technical features may be implemented in the form of program instructions that can be executed through various computer means and recorded in a computer-readable medium. The computer-readable medium may include program instructions, data files, data structures, and the like, alone or in combination. The program instructions recorded on the medium may be those specially designed and constructed for the embodiments or may be available to those skilled in the art of computer software. Examples of computer-readable media include magnetic media such as hard disks, floppy disks and magnetic tape; optical media such as CD-ROMs and DVDs; magnetic media such as floppy disks; Magneto-optical media, and hardware devices specifically configured to store and execute program instructions such as ROM, RAM, flash memory, and the like. Examples of program instructions include machine language code such as those produced by a compiler, as well as high-level language code that can be executed by a computer using an interpreter or the like. The hardware device may be configured to operate as one or more software modules to perform the operations of the embodiments, and vice versa.

As described above, the present invention has been described with reference to particular embodiments, such as specific elements, and specific embodiments and drawings. However, it should be understood that the present invention is not limited to the above- And various modifications and changes may be made thereto by those skilled in the art to which the present invention pertains. Accordingly, the spirit of the present invention should not be construed as being limited to the embodiments described, and all of the equivalents or equivalents of the claims, as well as the following claims, belong to the scope of the present invention .

Claims (15)

Calculating a relative reliability between the two users by using two users' ratings for at least one item and an average rating for the items;
Connecting an edge between the nodes where the relative reliability exists and creating a trust network having the nodes as a plurality of users including the two users; And
Predicting relative reliability between nodes for which the relative reliability does not exist, using the calculated relative reliability,
The step of predicting the relative reliability
Determining a shortest path between a first node and a second node in which the relative reliability does not exist; And
Estimating a relative reliability by applying a weight to the relative reliability of edges included in the shortest path,
And calculating a relative reliability for the recommendation system.
The method according to claim 1,
The step of calculating the relative reliability comprises:
The relative reliability is calculated using a difference between the ratings of the two users and a difference between the user's rating and the average rating
A Relative Reliability Calculation Method for Recommendation System.
3. The method of claim 2,
The average rating
The average rating of the users other than the two users
A Relative Reliability Calculation Method for Recommendation System.
The method according to claim 1,
The step of calculating the relative reliability comprises:
Calculating a first relative reliability for a second user of the first user; And
Calculating a second relative reliability for the first user of the second user
And calculating a relative reliability for the recommendation system.
5. The method of claim 4,
The edge
Wherein the first and second relative reliability indicators
A Relative Reliability Calculation Method for Recommendation System.
delete The method according to claim 1,
The weight
And is adjusted according to the distance between the first node and the edge
A Relative Reliability Calculation Method for Recommendation System.
The method according to claim 1,
The step of predicting the relative reliability
Applying the weight to the relative reliability of an edge that is below a critical distance from the first node
A Relative Reliability Calculation Method for Recommendation System.
Calculating a relative reliability between the two users by using two users' ratings for at least one item and an average rating for the items;
Connecting an edge between nodes having the calculated relative reliability to create a trust network having a plurality of users including the two users as the nodes;
Estimating a relative reliability among users whose relative reliability does not exist using the calculated relative reliability; And
Based on the relative reliability, recommending an item for a target user,
The step of predicting the relative reliability
Determining a shortest path between a first node and a second node in which the relative reliability does not exist; And
Estimating a relative reliability by applying a weight to the relative reliability of edges included in the shortest path,
Based item recommendation method.
10. The method of claim 9,
The step of calculating the relative reliability comprises:
The relative reliability is calculated using a difference between the ratings of the two users and a difference between the user's rating and the average rating
Trust - based item recommendation method.
delete delete 10. The method of claim 9,
The weight
And is adjusted according to the distance between the first node and the edge
Trust - based item recommendation method.
delete delete

Images